Research map / Reinforcement learning / Visual and representation RL
DoughNet: A Visual Predictive Model for Topological Manipulation of Deformable Objects
Dominik Bauer, Zhenjia Xu, Shuran Song
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| predict the next latent set directly instead of relative change (ablation) | DoughNet · Simulated test scenes, full sequence | VIoU: 73.1 -> 92.0 | Improved |
| supervise predictions directly in latent space instead of via reconstruction (ablation) | DoughNet · Simulated test scenes, full sequence | VIoU: 76.3 -> 92.0 | Improved |
| train on partial noisy synthetic observations instead of complete ones (ablation) | DoughNet · Simulated test scenes, single step | VIoU: 77.0 -> 94.3 | Improved |
| train two prediction steps (one from encoded state, one from own prediction) (ablation) | DoughNet · Simulated test scenes, full sequence | VIoU: 90.8 -> 92.0 | Improved |
| replace latent supervision with explicit reconstruction supervision (ablation) | DoughNet · Simulated test scenes, single step | VIoU: 94.3 -> 94.8 | No clear effect |
| use learned topology prediction instead of post-hoc static topology check | DoughNet · Simulated test scenes, final time step | AccC: 68.7 -> 92.3 | Improved |
| use learned model instead of MLS-MPM simulation with dynamic topology oracle | DoughNet · Simulated test scenes | fps: 0.2 -> 22.1 | Improved |
| use learned model instead of MLS-MPM at final time step | DoughNet · Simulated test scenes, final time step | CIoU: 79.4 -> 75.1 | Hurt |
| use object-level latent set model instead of GNN-based RoboCook | DoughNet · Simulated test scenes, full sequence | CIoU: 59.2 -> 90.5 | Improved |
| detect topological change by z-score of frame-to-frame latent distance | DoughNet latent space · Topological change detection | F1-score: 87.3% | Improved |
Closest papers by meaning
- DeformGen: Dynamics-Based Topology Augmentation for Deformable Manipulation Policy Learning (NeurIPS 2026 · similarity 0.89)
- Dynamic Environments with Deformable Objects (NeurIPS 2021 · similarity 0.89)
- DeformMaster: An Interactive Physics-Neural World Model for Deformable Objects from Videos (NeurIPS 2026 · similarity 0.87)
- Geometry-aware RL for Manipulation of Varying Shapes and Deformable Objects (ICLR 2025 · similarity 0.86)
- DynaRend: Learning 3D Dynamics via Masked Future Rendering for Robotic Manipulation (NeurIPS 2025 · similarity 0.85)
- UniGarmentManip: A Unified Framework for Category-Level Garment Manipulation via Dense Visual Correspondence (CVPR 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
